Snowflake CoCo is Snowflake's AI-powered coding agent. It is available in two interfaces: directly in Snowsight for web-based use, and as a command-line interface (CLI) for developers who prefer the terminal. This guide focuses on the CLI.

Snowflake CoCo CLI lets you query data, build applications, and manage Snowflake resources using plain English directly from your terminal. Behind the scenes it writes and runs SQL, orchestrates Snowflake-native skills, and connects to external tools through the Model Context Protocol (MCP).

In this guide you will install the CLI, connect it to a Snowflake account, and run a handful of queries to see what it can do.

A working Snowflake CoCo CLI environment connected to your Snowflake account, ready for day-to-day use.

Open a terminal and run the installer:

The script installs the cortex executable to ~/.local/bin and adds it to your PATH.

Open PowerShell and run:

The executable installs to %LOCALAPPDATA%\cortex and is added to your PATH automatically.

You should see the installed version number printed to the terminal.

Launch Snowflake CoCo CLI by typing cortex in your terminal. A setup wizard walks you through connecting to Snowflake.

The wizard presents two options:

Snowflake CoCo CLI supports two authentication methods:

You can generate a PAT from Snowsight or via SQL. See Using programmatic access tokens for authentication for details.

Once connected, Snowflake CoCo CLI drops you into an interactive session. Type:

This prints your active connection, role, warehouse, database, and schema.

With the session running, type a plain-English question:

Snowflake CoCo CLI translates your request into SQL, runs it against Snowflake, and returns the results directly in the terminal.

Try a few more requests to get a feel for what is possible:

Snowflake CoCo CLI displays its reasoning steps as it works. If it needs more information it will ask you a follow-up question.

When a query returns tabular data, press Ctrl+T to open the built-in table viewer. Inside the viewer you can scroll through results, cycle between tables with Tab, and press c to copy the query to your clipboard.

You can also run SQL directly using the /sql slash command:

To limit the number of rows displayed:

Prefix a fully qualified table name with # to pull its schema and sample rows into the conversation automatically:

Snowflake CoCo CLI fetches the column definitions and a sample of rows so it can answer questions about the table without you having to describe it.

Every conversation is saved automatically. You can pick up exactly where you left off.

This opens a list of previous sessions. Select one and press Enter, or use cortex --continue when launching to jump straight into the most recent session.

Made a wrong turn? Roll back to an earlier point in the conversation:

Want to try a different approach without losing your current progress? Fork the session:

This creates a new session branched from the current point.

To limit what Snowflake CoCo CLI can do in the current session, for example read-only access or only specific roles, apply a Restricted Session Scope:

Select Restrict this session, then apply an existing scope or choose read-only access. Run /guardrails status to see the active restriction. You can also apply a named scope at launch:

Once a restriction is active you can switch to another scope, but you can't remove or relax it in the current session. Start a new session to run without it.

Congratulations! You've successfully installed Snowflake CoCo CLI, connected it to your Snowflake account, and run your first natural-language queries from the terminal.

From here you can:

Documentation:

Guides:

Additional Reading: